Solution Brief Data Center | Hybrid Cloud
Get Fast Insights with a VMware Hybrid Cloud and Intel® Optane™ Technology
VMware Cloud Foundation, powered by Intel® technology, delivers high performance as well as simplified, user-friendly deployment and management of critical business workloads
Executive Summary The speed and ease with which companies are able to gain business insights can make the difference between having the competitive edge and simply trying to keep up. The digital landscape is changing rapidly, and traditional hosting applications and services can no longer deliver the necessary innovations at the speed businesses require. And with the cost of many new technologies, businesses Solution Benefits are faced with difficult challenges. • Fast deployment. Quickly launch Rapid software and hardware provisioning, deployment, and maintenance, as well database processing, analytics, as application development, testing, and delivery can help companies get ahead AI, and machine-learning and stay ahead. VMware Cloud Foundation (VCF), powered by Intel® technology, workloads on a platform that offers a solution with a simplified hybrid cloud platform for easier management scales to future needs. of virtual machines (VMs) and containers (see Figure 1). Now businesses can • Easy management. Migrate operate workloads, such as data analytics, artificial intelligence (AI), and machine workloads between public learning (ML)—foundational applications for predicting trends—across private clouds and private data centers and public clouds with excellent performance and reliability without modifying without modifying applications node configuration or application refactoring. The platform is designed to scale or node configurations. as businesses grow and workload demands increase. Integrating Intel® Optane™ persistent memory (PMem) into VCF clusters improves transactions per minute • Excellent performance. Whether (TPM) application performance, including data warehousing performance, by an running VMs or orchestrating average of 3.34x.1 containers, the latest Intel® tech- nology provides the throughput and responsiveness critical for today’s data-intensive workloads. VMware Cloud Foundation ware S anager ublic loud
ri ate loud
ware Sphere ware Reali e Suite ware SA ware S
Server Hardware
2nd en ntel eon ntel SS s ntel ptane ntel thernet Scalable rocessor ersistent emor Adapter Author Patryk Wolsza Figure 1. VMware Cloud Foundation, powered by Intel® technology, is a hybrid vExpert, Cloud Solutions Architect cloud solution that retains the benefits of on-premises infrastructure while Intel Cloud & Enterprise Solutions providing the agility of a public cloud. Solution Brief | Get Fast Insights with a VMware Hybrid Cloud and Intel® Optane™ Technology 2
Business Challenge: Traditional Workload Solution Value: Scalable Processing Now Management is Cumbersome and in the Future The digital landscape is constantly changing, and competing To demonstrate the value of deploying modern workloads for market share requires rapidly identifying trends in on VCF using the latest Intel technologies, Intel recently enterprise data. But to do so, companies must accelerate conducted performance tests for both data warehousing workload deployment and simplify the overall management. and machine-learning workloads. The tests results showed The addition of end-user self-service solutions has further that software optimized for Intel® hardware, plus innovations reduced expected time-to-market. Modern technology such as Intel® Optane™ persistent memory (PMem) can boost solutions must transcend the divide between hardware application performance. For detailed information about and software, delivering an end-to-end solution that testing and results, as well as a bill of materials, see the full encompasses hardware provisioning, deployment, and Reference Architecture. maintenance, as well as application development, testing, and delivery. Data Warehousing Benefits from Intel® Optane™ PMem With hybrid cloud infrastructure, companies can retain the Testing showed that Microsoft SQL Server performance benefits, security, and control of a familiar on-premises improves when Intel Optane PMem in App Direct Mode is infrastructure while gaining the flexibility and instant added. The test compared the number of transactions per availability of the public cloud. VMware Cloud Foundation minute (TPM) achieved by a “Base cluster” without Intel (VCF) with the latest Intel® technology provides the flexibility Optane PMem to the TPM achieved by a “Plus cluster” with and agility companies need for everything from database Intel Optane PMem and a high-performance Intel® Xeon® processing and analytics to artificial intelligence (AI) and Platinum processor. The Plus cluster provided an average machine learning (ML), with scalability for future growth. of 3.34x improvement in TPM and scaled linearly with the addition of VMs (see Figure 3).2
Flexible Hybrid Cloud Platform for a Transactions Per Minute Comparison2 ase s. lus onfiguration Variety of Modern Workloads HIGHER IS BETTER
Many important enterprise workloads fall into two lus broad areas: data warehousing and AI/ML. Most of these ase applications, like Microsoft SQL Server, Oracle Database, and DataRobot, can run as VMs next to other workloads 3.34x in Kubernetes-based containers on the same platform. PERFORMANCE Containers are treated as first-level citizens. VCF supports lus Con i ur tion Incre se both approaches (see Figure 2), so that enterprises can deploy a workload in the way that makes most business sense.
VCF enables businesses to provision infrastructure from a Normalized TPM pool of virtualized resources on premises, at the edge, or in the cloud. Modern apps can take advantage of built-in automation and orchestration to simplify management of VM- or container-based workloads. And when the business 2 2 grows, VCF deployments can scale as necessary. 2 Number of VMs Figure 3. The Plus cluster achieved an average of 3.34x more Data Warehousing s Machine earning Workloads TPM compared to the Base cluster and is highly scalable. A A A A A TA R TA R S S S S S TA R TA R
Software efined ata enter: ware loud oundation with T i
Software efined etworking: ware S
Software efined Automation: ware Reali e Suite
Software efined Storage: ware SA ternal atastore
anagement: ware enter Hypervisor S i S i 2 ware S i S i S i n
Figure 2. VMware Cloud Foundation can run a wide variety of data analytics, AI, and machine-learning workloads. Solution Brief | Get Fast Insights with a VMware Hybrid Cloud and Intel® Optane™ Technology 3
AI/Analytics: Deep-Learning Inference Solution Architecture: High Performance, In recent tests, Intel benchmarked two image classification User-Friendly, Scalable Processing topologies, comparing the standard version of TensorFlow versus the TensorFlow distribution from the Deep Learning User-friendly VCF for hybrid cloud computing uses Intel Reference Stack with pretrained models from Model Zoo for technology to help achieve high performance and scalability Intel® Architecture. The test consisted of two experiments, (see Figure 6 for the solution architecture). The solution one with a fat VM and another with a Tanzu Kubernetes includes the following technologies: Grid Integrated (TKGi) instance. The results demonstrate the benefits of software optimized for Intel architecture for Intel® Technology both topologies we tested. • Intel Optane PMem helps improve server performance by For the fat VM test, using the Deep Learning Reference supplying large amounts of persistent storage with low-latency Stack version of TensorFlow improved the performance access. It provides capacity that is unavailable with traditional of the Inception v3 topology for the Base configuration by DRAM modules, and operating modes can be configured 2.4x, and improved the throughput of the Plus configuration using the platform BIOS or memory management tool. by 3x (see Figure 4).3 • 2nd Generation Intel Xeon Scalable processors are workload
3 optimized for analytics, AI, and in-memory database Inception v3 Performance Comparison workloads to process ever-increasing amounts of data. Standard Tensor low s. eep earning Reference Stack HIGHER IS BETTER . • Intel® Optane™ SSDs in the cache layer and large-capacity NVMe-based 3D NAND SSDs in the capacity layer help VMware vSAN optimize performance. 3x • Intel® Volume Management Device (Intel® VMD) IMPROVEMENT with the eep technology for NVMe drivers enables serviceability of earning Reference 2.4x Stack NVMe-based SSDs by supporting hot swap replacement IMPROVEMENT from the PCle bus without shutting down. with the eep earning Reference • accelerates performance for Stack Intel® Ethernet 700 Series VMware hybrid cloud platform to help meet enterprise-
frames per second level requirements for data resiliency, service reliability,
Normalized Performance aseline aseline for Standard for Standard and ease-of-provisioning. Tensor low Tensor low ase onfiguration lus onfiguration . Cloud Infrastructure Base Configuration Plus Configuration • VMware Cloud Foundation offers a simplified path to Figure 4. Software optimized for Intel® architecture accrues hybrid cloud computing through an integrated software significant performance benefits. platform for both private and public cloud environments. It includes a complete set of software-defined services For the Kubernetes cluster provisioned by TKGi, we for compute, storage, network, and security, as well as measured throughput scaling for the Deep Learning application-focused cloud management capabilities. Reference Stack container by running the benchmark on one to six workers in parallel for the deep-learning workloads. • VMware Software-Defined Data Center (SDDC) Manager We observed improvement in throughput with additional manages VCF system bring-up, as well as creates and jobs running in parallel, demonstrating the effectiveness manages workload domains and performance lifecycles. of ESXi scheduler (see Figure 5).4 It also manages logical and physical resources. • VMware vSphere extends virtualization to storage Inception v3 Throughput Scaling and network services, and automated, policy-based for the Plus Configuration provisioning and management.
HIGHER IS BETTER . • VMware NSX-T Data Center (formerly NSX-T) is a network . virtualization platform that enables software-defined . 2 virtualization by working like a network hypervisor. It . reproduces Layer 2-7 network services: routing, switching, 2. access control, firewalls, quality of service (QoS), and . Dynamic Host Configuration Protocol (DHCP) in software. • VMware Cloud (on AWS) is a hybrid cloud solution that . allows easy extension, migration, and modernization of frames per second Relative Throughput applications. It also protects applications in the public cloud.
. • TKGi is a container solution with advanced networking, a 2 private container registry, and life cycle management. It Number of Workers simplifies deployment and operation of Kubernetes clusters Figure 5. Optimized software, combined with high-performance so you can run and manage containers at scale on private hardware, results in excellent scalability. and public clouds. Solution Brief | Get Fast Insights with a VMware Hybrid Cloud and Intel® Optane™ Technology 4
Analytics AI VMware Cloud on AWS ntel Architecture ptimi ed uilding locks achine earning ata Warehousing ramework ware ntel istribution for thon ntel loud lock ware Software efined ataRobot Spanning ata enter anager eep earning Reference Stack ntel oS atabases S atabases
ware Sphere luster Cloud Infrastructure: VMware Cloud Foundation 3. ware S T ware SA ware Sphere luster ware Software efined ware Reali e Suite ware S T ware ware SA Sphere . luster ware S i per isor ata enter anager ware S T ware SA ware S i per isor ware enter ware Reali e Suite brid ink Tan u ubernetes rid integrated ware Reali e Suite ode ware Sphere ware enter Ser er T i rchestration ware enter Ser er
Bare Metal Intel® Data Center Blocks For Cloud VMware (vSAN ReadyNodes) Infrastructure onsistent igh erformance Architecture ased on ntel Technolog Ama on 2 ntel ptane SS 2nd en ntel eon Series ntel thernet Scalable rocessors are etal nstances Series e ased SS old 2 2 i .metal A Series latinum 2 r .metal
Figure 6. VMware Cloud Foundation with Intel® technology offers an end-to-end solution to meet today’s needs, as well as future growth.
Conclusion Companies are increasingly pressed to gain faster insights from their data to remain competitive. Traditional hosting Learn More applications and services cannot deliver the necessary innovations at the speed businesses require today and as You may also find the following resources useful: they grow in the future. More than ever, businesses are • Building a VMware Hybrid Cloud Platform Reference turning to modern applications, such as analytics, AI, and Architecture ML to overcome these difficult challenges. And modern • Intel® Data Center Blocks for Cloud – vSAN ReadyNodes applications require dynamic cloud-native environments that • Intel® Optane™ persistent memory home page offer ease of use, scalability, rapid hardware provisioning, • Intel Optane Persistent Memory – Virtualized Performance deployment, and maintenance, as well as application • Deep Learning Reference Stack development, testing, and delivery. • 2nd Generation Intel® Xeon® Scalable processors VCF powered by Intel architecture provides a simplified • Intel® Select Solutions for VMware vSAN ReadyNodes hybrid cloud platform for easier management of VMs and • Intel® Optane™ Solid State Drives containers that also allows AI/ML workloads to coexist with • Intel® Solid State Drives Data Center Family legacy workloads on the same platform. Businesses can now • Intel® Framework Optimizations operate workloads across private and public clouds with excellent performance and reliability (without modifying • VMware vSAN node configurations) in one scalable solution. • VMware Cloud Foundation • VMware Cloud on AWS Find the solution that is right for your organization. Visit intel.com/optane or contact your Intel representative. Solution Brief | Get Fast Insights with a VMware Hybrid Cloud and Intel® Optane™ Technology 5
Solution Provided By:
1 Testing by Intel, June 15-23, 2020. Each cluster (Base, Plus, and the Management cluster that were used for workload generation) consists of four machines. All machines are identical within the given cluster. On top of those clusters the VMs are placed: five VMs per node for Base and eight VMs per node for Plus. We had four scenarios for each environment (Base and Plus). The first scenario used only one node in the cluster, the second scenario used two, and so on, up until all four nodes were used. Each scenario was repeated three times; the results given here are the average values for these three runs for each scenario. Base 4-node cluster configuration: Intel® Xeon® 6248 processor (2.5 GHz, 20 cores); 384 GB DRAM (12x 32 GB 2933 MHz); Intel® Hyper-Threading Technology ON; Intel® Turbo Boost Technology ON; Storage: VMware vSAN, disk group: 1x Intel® Optane™ SSD DC P4800X 375 GB and 2x Intel® SSD DC P4510 2 TB, two disk groups per node; BIOS = 02.01.0010; microcode = 0x0500002c; OS = VMware ESXi 6.7.015160138; 1x Intel® Ethernet Adapter XXV710-DA2. Base VM configuration: 8 vCPUs; 44 GB vMemory; vDisks: 100 GB – OS drive (LSI controller), 4x 18 GB – Data DB (1 SCSI controller), 4 files per disk, 2x 200 MB – Temp DB (1 SCSI controller), 200 MB – Temp log (same SCSI controller as User transaction log), 10 GB – User transaction log (1 SCSI controller); vNetwork = VMXNET3; OS = Windows Server 2019 Datacenter 17763.rs5.180914-1434; Database = MSSQL Server 2019 15.0.2070.41. Plus 4-node cluster configuration: Intel® Xeon® 8268 processor (2.9 GHz, 24 cores); 384 GB DRAM (12x 32 GB 2933 MHz) plus 1,536 GB Intel® Optane™ persistent memory (12x 128 GB); Intel Hyper-Threading Technology ON; Intel Turbo Boost Technology ON; Storage: VMware vSAN, disk group: 1x Intel Optane SSD DC P4800X 375 GB and 2x Intel SSD DC P4510 2 TB, two disk groups per node; BIOS = 02.01.0010; microcode = 0x0500002c; OS = VMware ESXi 6.7.0 15160138; 1x Intel Ethernet Adapter XXV710-DA2; PMem used a 2-2-2 configuration in App Direct Mode; PMem firmware version = 01.02.00.5417. Plus VM configuration: 8 vCPUs; 44 GB vMemory; vDisks: 100 GB – OS drive (LSI controller), 4x 18 GB Data DB (NVDIMM), 4 files per disk, 2x 200 MB Temp DB (NVDIMM non-DAX volume), 200 MB Temp log (NVDIMM non-DAX volume), 10 GB User Transaction log (NVDIMM non-DAX volume); vNetwork = VMXNET3; OS = Windows Server 2019 Datacenter 17763.rs5.180914- 1434; Database = MSSQL Server 2019 15.0.2070.41. 2 See endnote 1. 3 Testing by Intel, June 15-23, 2020. Workload: Inception V3 topology at fp32 precision. Default TensorFlow tensorflow/tensorflow:1.15.0-py3; optimized TensorFlow: clearlinux/stacks- dlrs-mkl:v0.5.0. Base 4-node cluster configuration: Intel® Xeon® 6248 processor (2.5 GHz, 20 cores); 384 GB DRAM (12x 32 GB 2933 MHz); Intel® Hyper-Threading Technology ON; Intel® Turbo Boost Technology ON; Storage: VMware vSAN, disk group: 1x Intel® Optane™ SSD DC P4800X 375 GB and 2x Intel® SSD DC P4510 2 TB, two disk groups per node; BIOS = 02.01.0010; microcode = 0x0500002c; OS = VMware ESXi 6.7.015160138; 1x Intel® Ethernet Adapter XXV710-DA2. Base Fat VM configuration: 80 vCPUs; OS = CentOS Linux release 8.1.1911; kernel = 4.18.0-147.5.1.el8_1.x86_64; other SW: VMware vSphere 6.7, VMware Cloud Foundation 3.9, VMware ESXi hypervisor 6.7 update 3. Normalized Base Results: 2.4X more throughput using the Deep Learning Reference Stack version of TensorFlow, compared to the default TensorFlow performance. Plus 4-node cluster configuration: Intel Xeon 8268 processor (2.9 GHz, 24 cores); 384 GB DRAM (12x 32 GB 2933 MHz) plus 1,536 GB Intel Optane persistent memory (12x 128 GB); Intel Hyper-Threading Technology ON; Intel Turbo Boost Technology ON; Storage: VMware vSAN, disk group: 1x Intel Optane SSD DC P4800X 375 GB and 2x Intel SSD DC P4510 2 TB, two disk groups per node; BIOS = 02.01.0010; microcode = 0x0500002c; OS = VMware ESXi 6.7.0 15160138; 1x Intel Ethernet Adapter XXV710-DA2; PMem used a 2-2-2 configuration in App Direct Mode; PMem firmware version = 01.02.00.5417. Plus Fat VM configuration: 96 vCPUs; otherwise, identical to the Base Fat VM configuration. Normalized Plus Results: 3X more throughput using the Deep Learning Reference Stack version of TensorFlow, compared to the default TensorFlow performance. 4 Testing by Intel, June 15-23, 2020. Workload: Inception V3 topology at fp32 precision. 6 worker VMs with 16 vCPU each. See endnote 3 for physical infrastructure configuration. Default TensorFlow: tensorflow/tensorflow:1.15.0-py3; optimized TensorFlow: clearlinux/stacks-dlrs-mkl:v0.5.0. Worker Node configuration: OS = Ubuntu 16.04.6 LTS; kernel = 4.15.0-55-generic; other SW: VMware vSphere 6.7, VMware Cloud Foundation 3.9, VMware ESXi hypervisor 6.7 update 3. Intel technologies may require enabled hardware, software or service activation. No product or component can be absolutely secure. Your costs and results may vary. Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more complete information visit intel.com/benchmarks. Intel’s compilers may or may not optimize to the same degree for non-Intel microprocessors for optimizations that are not unique to Intel microprocessors. These optimizations include SSE2, SSE3, and SSSE3 instruction sets and other optimizations. Intel does not guarantee the availability, functionality, or effectiveness of any optimization on microprocessors not manufactured by Intel. Microprocessor-dependent optimizations in this product are intended for use with Intel microprocessors. Certain optimizations not specific to Intel microarchitecture are reserved for Intel microprocessors. Please refer to the applicable product User and Reference Guides for more information regarding the specific instruction sets covered by this notice. Intel does not control or audit third-party data. You should consult other sources to evaluate accuracy. Performance results are based on testing as of dates shown in configurations and may not reflect all publicly available updates. See backup for configuration details. No product or component can be absolutely secure. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation in the U.S. and/or other countries. Other names and brands may be claimed as the property of others. © Intel Corporation 0920/SALL/KC/PDF 344029-001US